Census Income Prediction – Data-Driven Income Classification (July 2025)

🔑 Skills Demonstrated

• Supervised machine learning using scikit-learn
• Logistic Regression and Decision Tree model implementation
• Hyperparameter tuning with GridSearchCV
• Data preprocessing with pandas, scaling with StandardScaler
• Model evaluation using accuracy, precision, recall, and classification reports
• Understanding overfitting vs. generalization through performance tradeoffs

---

🧠 Project Summary

This project uses 32,000+ anonymized U.S. Census records to build a machine learning model that predicts whether an individual earns more than $50K annually. The goal was to evaluate and compare different classification algorithms and tune them to achieve strong performance metrics while understanding the balance between underfitting and overfitting.

---

🧪 Approach

1. Data Preparation

   • Loaded data from censusData.csv
   • Converted target column income_binary to 0/1 (<=50K → 0, >50K → 1)
   • Scaled features using StandardScaler for logistic regression

2. Modeling

   • Trained a baseline LogisticRegression model (with max_iter=1000)
   • Implemented a Decision Tree classifier and tested various depths and leaf sizes
   • Used GridSearchCV to find optimal tree parameters (max_depth=10, min_samples_leaf=1, criterion='gini')

3. Evaluation

   • Achieved ~85% accuracy on both models
   • Logistic Regression: Recall ≈ 58.5%, Precision ≈ 71.2%
   • Decision Tree (after tuning): Recall ≈ 57.9%, Precision ≈ 73.3%

4. Insights

   • Logistic Regression provided strong overall performance with simpler structure
   • Decision Tree offered more interpretability and competitive precision after tuning
   • Grid search was key in avoiding overfitting and improving generalization

---

📊 Final Results

Model	Accuracy	Precision	Recall
Logistic Regression	84.7%	71.2%	58.5%
Decision Tree (tuned)	85.1%	73.3%	57.9%

---

📁 Files

• DefineAndSolveMLProblem.ipynb – Full notebook with code and evaluation
• censusData.csv – Dataset used for training and testing

---

🤔 Reflections

This project helped solidify my understanding of:

• Model selection tradeoffs (simplicity vs flexibility)
• Evaluating models beyond accuracy using recall and precision
• Preventing overfitting through hyperparameter tuning
• Data preprocessing techniques essential to model performance